In recent years, scientists have been paying more and more attention on extracting features from the radiated noise of underwater targets. Thus, enriching the feature reserve of underwater targets is quite significant for scientists in order to detect and study them. The paper presents an algorithm of feature extraction, which focuses on the MFCC feature coefficients of underwater targets. Mel Frequency Cepstral Coefficients (MFCCs) are based on the nonlinear frequency feature of human ears. In essence, MFCC works via selecting energy in different frequency bands as the feature of target. Because of its outstanding performance in expressing speech spectrum at low frequency, since it is a good simulation of human auditory sensation, it has been one of the most important features used in speaker recognition systems. However, whether it is applicable in the case of expressing the features of underwater targets was still unclear. According to the result of a series of correlative experiments and researches, scientists found that the principle of distinguishing different underwater radiated noises by sonarman is the same as voice recognition by human ears. Meanwhile, the method of extracting MFCC has some advantages. For example, noises at low frequencies (in the audible range), which are the main sources of radiated noises ships and submarines, can propagate for a long distance. Fortunately, the method of extracting MFCC is robust to resist the disturbance of background noise at that frequency band. At the same time, seas and oceans always have chaotic background noise. The acoustic processes underwater are usually very complicated and nonlinear, and therefore requiring a proper nonlinear principle. Thus, MFCC can be applied to feature extraction of underwater radiated noises. In this paper, the radiated noises of different marine lifes (whales, sea lions and dolphins ), divers, boats and ships are all researched. Their MFCC feature coefficients are extracted and compared. The results show that different targets have clear differences in MFCC feature coefficients. Therefore, MFCC can be an effective feature for extraction and recognition.
[1]
Zhu Jin.
Feature extraction of ship-radiated noise based on Mel frequency cepstrum coefficients
,
2004
.
[2]
J. Sabatier,et al.
Vibration and sound signatures of human footsteps in buildings.
,
2006,
The Journal of the Acoustical Society of America.
[3]
Alpay Koç,et al.
ACOUSTIC FEATURE ANALYSIS FOR ROBUST SPEECH RECOGNITION
,
2002
.
[4]
Douglas A. Reynolds,et al.
Speaker identification and verification using Gaussian mixture speaker models
,
1995,
Speech Commun..
[5]
Shi,et al.
Application of auditory spectrum-based features into acoustic target recognition
,
2010
.
[6]
Sridha Sridharan,et al.
Feature warping for robust speaker verification
,
2001,
Odyssey.
[7]
Zhang Weihao.
Experiment and study for passive acoustic detection of underwater divers
,
2012
.
[8]
Anthony J. Bonner,et al.
Machine analysis of acoustical signals
,
1983,
Pattern Recognit..